Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 http://www.biomedcentral.com/1755-8794/4/80

RESEARCHARTICLE Open Access Analysis of BMP4 and BMP7 signaling in breast cancer cells unveils time-dependent transcription patterns and highlights a common synexpression group of Alejandra Rodriguez-Martinez1†, Emma-Leena Alarmo1†, Lilli Saarinen2, Johanna Ketolainen1, Kari Nousiainen2, Sampsa Hautaniemi2 and Anne Kallioniemi1*

Abstract Background: Bone morphogenetic (BMPs) are members of the TGF-beta superfamily of growth factors. They are known for their roles in regulation of osteogenesis and developmental processes and, in recent years, evidence has accumulated of their crucial functions in tumor biology. BMP4 and BMP7, in particular, have been implicated in breast cancer. However, little is known about BMP target genes in the context of tumor. We explored the effects of BMP4 and BMP7 treatment on global transcription in seven breast cancer cell lines during a 6- point time series, using a whole-genome oligo microarray. Data analysis included hierarchical clustering of differentially expressed genes, enrichment analyses and model based clustering of temporal data. Results: Both ligands had a strong effect on gene expression, although the response to BMP4 treatment was more pronounced. The cellular functions most strongly affected by BMP signaling were regulation of transcription and development. The observed transcriptional response, as well as its functional outcome, followed a temporal sequence, with regulation of gene expression and signal transduction leading to changes in metabolism and cell proliferation. Hierarchical clustering revealed distinct differences in the response of individual cell lines to BMPs, but also highlighted a synexpression group of genes for both ligands. Interestingly, the majority of the genes within these synexpression groups were shared by the two ligands, probably representing the core molecular responses common to BMP4 and BMP7 signaling pathways. Conclusions: All in all, we show that BMP signaling has a remarkable effect on gene transcription in breast cancer cells and that the functions affected follow a logical temporal pattern. Our results also uncover components of the common cellular transcriptional response to BMP4 and BMP7. Most importantly, this study provides a list of potential novel BMP target genes relevant in breast cancer. Keywords: bone morphogenetic , breast cancer, BMP4, BMP7, expression microarray

Background regulate transcription of target genes by signaling Bone morphogenetic proteins (BMPs) are extracellular through type I and II transmembrane serine-threonine ligand molecules that belong to the transforming growth receptors. Binding of the ligand to the type II receptor factor b (TGF-b) superfamily. To date, 21 members of elicits phosphorylation of the type I receptor, which, as the human BMP family have been identified [1]. BMPs a result, is able to phosphorylate other molecules and transmit the signal. In the canonical BMP pathway, the * Correspondence: [email protected] type I receptor phosphorylates receptor-regulated † Contributed equally SMAD (homologue of Drosophila Mothers Against Dec- 1Laboratory of Cancer Genetics, Institute of Biomedical Technology, University of Tampere and Centre for Laboratory Medicine, Tampere apentaplegic) proteins (R-SMADs, SMAD-1/5/8), which University Hospital, Finland then bind to the common mediator SMAD4; the Full list of author information is available at the end of the article

© 2011 Rodriguez-Martinez et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 2 of 16 http://www.biomedcentral.com/1755-8794/4/80

resulting SMAD complex translocates to the nucleus to time series, using a genome-wide approach. We charac- regulate transcription of target genes [1]. The signals terized the transcriptional response of breast cancer generated by BMPs in the cell membrane may be also cells to BMP signaling in an analysis that included a transferred into the cell via ERK, JNK and p38 mitogen- temporal dimension and the comparison of different cell activated protein kinases (MAPK) [2,3]. Moreover, there lines and two BMP ligands. Finally and most impor- is crosstalk between BMP signaling and other cellular tantly, we report novel potential BMP target genes rele- signaling cascades, such as the Wnt, JAK/STAT and vant in breast cancer. Notch pathways [4-6]. BMPs were first identified as inducers of ectopic bone Methods formation in vivo [7] but were later found to be crucial Breast cancer cell lines multifunctional regulators of development [8]. During Seven breast cancer cell lines (HCC1954, MDA-MB-361, the last decade, the role of BMPs in cancer development ZR-75-30, HCC1419, SK-BR-3, MDA-MB-231 and T- has gained increasing interest [9-11]. The importance of 47D) were purchased from the American Type Culture BMP4 and BMP7 in breast cancer was highlighted in a Collection (ATCC, Manassas, VA) and cultured accord- survey of seven BMPs: these two ligands had the highest ing to the recommended conditions except for MDA- expression levels and were the most frequently MB-231 and T-47D, for which the concentration of FBS expressed among 22 cell lines and 39 primary tumor in culture media were 1% and 5%, respectively. samples [12]. The expression of BMP4 and BMP7 in breast cancer also has been demonstrated in several BMP4 and BMP7 treatments other reports [13-17]. Interestingly, BMP7 protein Recombinant human BMP4 and BMP7 proteins were expression in primary breast tumors has been associated purchased from R&D Systems (Minneapolis, MN). with accelerated bone metastasis formation and served Three cell lines (HCC1954, MDA-MB-361 and ZR-75- as an independent prognostic factor for early bone 30) were treated with both BMP4 (100 ng/ml) and metastasis in a study based on a set of 409 patient sam- BMP7 (50 ng/ml) separately. HCC1419 and SK-BR-3 ples [15] though, with a smaller set of 67 patient sam- cell lines received only BMP4 treatment (100 ng/ml), ples, this association was not established [18]. while MDA-MB-231 and T-47D were treated only with The functional significance of BMP4 and BMP7 in BMP7 (50 ng/ml). Cells were seeded on 24-well plates, breast cancer has been studied predominantly through allowed to adhere for 24 h, and treated with the BMP the use of in vitro models. BMP4 was shown to inhibit ligand or vehicle for 30 min, 1 h, 3 h, 6 h, 12 h and 24 cell proliferation in a panel of breast cancer cell lines by h (Figure 1A). Experiments were performed in triplicate inducing a G1 cell cycle arrest [14]. The effects of exo- and collected cells were pooled. genous BMP4 on breast cancer cell migration and inva- sion have also been studied. For the most part, the data Microarrays suggest promotion of these cellular abilities by BMP4 in Total RNA was extracted using the RNeasy Mini Kit several breast cancer cell lines and in normal breast (Qiagen, Valencia, CA) and the quality of RNA was vali- epithelial cells [14,19], while a study in which only dated using the Agilent RNA 6000 Nano Kit (Agilent MDA-MB-231 cells were analyzed reported the opposite Technologies, Palo Alto, CA, USA). Total RNA (500 ng) phenotype [20]. For BMP7, the results from different was used to generate fluorescent Cy-3-(vehicle treated reports and different cell lines are more variable. In cells) or Cy-5-labeled cRNA (BMP4- or BMP7-treated vitro examination of BMP7 manipulation have revealed cells) using the Agilent Low RNA Input Fluorescence cell line-specific effects on cell proliferation, migration Linear Amplification Kit (Agilent Technologies). The and invasion; BMP7 induces all of these parameters in labeled cRNAs were hybridized to the 44 K Whole MDA-MB-231 cells and inhibits cellular proliferation in oligo microarrays (Agilent Technolo- several other cell lines [21]. In opposition, in an in vivo gies) according to the manufacturer’s protocol. Microar- xenograft mouse model of MDA-MB-231 cells, BMP7 ray slides were scanned (Agilent Microarray Scanner) reduced tumor growth as well as the formation and after hybridization, and data were extracted using the growth of bone metastases [18]. Feature Extraction software, version A.7.5.1 (Agilent In spite of the many years since the discovery of BMPs Technologies). The microarray data has been submitted and being currently a very active topic in cancer to the GEO database (accession number GSE31605). research, little is known about their target genes in tumor conditions. The present study was designed to Data analysis gainknowledgeinthistopic,byexploringtheeffectsof Themicroarraydatawerefirstsubjectedtolinearnor- BMP4 and BMP7 signaling on gene transcription in malization to allow comparison between arrays. All seven breast cancer cell lines and throughout a 6-point probes were compared to the reference sequence using Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 3 of 16 http://www.biomedcentral.com/1755-8794/4/80

BMP4 BMP7 DEP number by cell line A Vehicle B 100ng/ml 50ng/ml 9000 8000 Treatment 7000 6000 5000 BMP4 4000 BMP7 HCC1419 MDA-MB-361 MDA-MB-231 3000 2000 Cell HCC1954 lines SK-BR-3 T-47D Number of DEPs 1000 ZR-75-30 0 9 3 1 4 1 0 1 - 3 D 5 6 3 4 R 2 7 9 3 - 1 - -4 1 - 5 -B B T B 7 C K C - C S -M C -M R H A H A Z Sample collection 30min 1h 3h 6h 12h 24h D D M M

C DEP number by time point

4500 BMP4 HCC1419 4000 BMP4 HCC1954 3500 BMP4 MDA-MB-361 3000 BMP4 SK-BR-3 2500 BMP4 ZR-75-30 2000 BMP7 HCC1954 1500 BMP7 MDA-MB-231 Number of DEPs 1000 BMP7 MDA-MB-361 500 BMP7 T-47D 0 BMP7 ZR-75-30 30min 1h 3h 6h 12h 24h

Figure 1 Experimental workflow and numbers of differentially expressed probes (DEPs) resulting from BMP4 or BMP7 treatment.(A) Seven breast cancer cell lines were cultured on 24-well plates, allowed to adhere for 24 h, and treated with the BMP ligand or vehicle for 30 min, 1 h, 3 h, 6 h, 12 h, and 24 h. Experiments were performed in triplicate, and collected cells were pooled. (B) The expression data from each cell line were individually filtered according to the following criteria: differential expression of at least 2-fold at a minimum of one time point. The number of DEPs per cell line is represented. (C) The expression data from individual time points of every cell line were filtered according to a 2-fold cutoff in expression change. The number of DEPs per time point is shown.

BLAST (v.2.2.23). Ensembl IDs for the probes were analysis. An event refers to any time point of any cell obtained by examining the probe’s genomic location. line, resulting in a maximum number of 30 events (5 The Ensembl Homo sapiens database version 60.37e was cell lines and 6 time points per cell line). The data sets used. This process resulted in the annotation of 84% of produced by general filtering were further hand-anno- all the probes. A total of 66% of the probes mapped tated to reduce the number of probes with multiple uniquely to genes and 16% mapped to multiple genes. annotations. Uniquely annotated probes are designated In order to determine differentially-expressed genes, hereafter as “genes, “ whereas the terms “probe” and expression data were subjected to three types of filter- “genetic element” refer to multiple annotated probes or ing: cell line-specific, time point-specific and general fil- any data including them. The gene lists resulting from tering. Cell line-specific filtering was done separately for general filtering were ranked according to the number each cell line, following the criteria of a differential of events in which they showed regulation. Furthermore, expression of at least 2-fold in a minimum of one time all the probes derived from general filtering were sub- point.Inthetimepoint-specificfiltering,datafrom jected to hierarchical clustering using correlation each time point were independently filtered according metrics, agglomerative strategy and average linkage to a 2-fold expression change cutoff. General filtering method. was performed on all the data from all the cell lines Enrichments of gene ontology (GO) terms were per- together (separately for BMP4 and BMP7) following the formed on several data sets applying Fisher’sexacttest next criteria: probes with a differential expression of at and using all genes present on the microarray as a refer- least3-foldinatleastthree events and/or 2-fold in at ence [22]. In all the GO enrichment analyses, only least four events were considered for subsequent probes with unique annotation were used. Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 4 of 16 http://www.biomedcentral.com/1755-8794/4/80

A model-based clustering method [23] that allows the expression data according to the following criteria: fold finding of clusters of genes with similar expression pro- change (FC) ≥ +/-3 in at least 3 events and/or FC ≥ files was performed using MCLUST R package [24]. +/-2 in at least 4 events. This resulted in the identifica- This method was applied to data resulting from cell tion of 2, 421 and 1, 263 differentially expressed gene line-specific filtering. Datawerelog2-transformedand elements (1, 678 and 905 uniquely annotated probes) scaled to unit length. In the model-based clustering for BMP4 and BMP7 experiments respectively, further method, the clusters are considered to be groups dis- evidencing a more prominent effect of BMP4 than playing multivariate distributions. Several models were BMP7 on gene transcription. fitted to the data. The selections of the best model and Unsupervised hierarchical clustering on the data sets the number of clusters were made based upon maximiz- resulting from general filtering (BMP4 and BMP7 sepa- ing Bayesian Information Criterion (BIC) values for the rately) revealed that the samples originating from a par- specific model and number of clusters that best repre- ticular cell line mainly clustered together (Figure 2), sented the data. The data analyses were performed suggesting considerable variation in the response of using the Anduril data analysis framework [25] and R individual cell lines to BMPs. The most obvious exam- [26]. ples are MDA-MB-361, ZR-75-30, and HCC1419 (for BMP4) as well as MDA-MB-361 and HCC1954 (for Results BMP7). Clustering according to time point was an The aim of this study was to uncover the transcriptional uncommon phenomenon, but it was observed for the responses of BMP4 and BMP7 signaling in breast can- samples derived from MDA-MB-231, ZR-75-30, cer. To this end, we selected breast cancer cell lines HCC1954 and T-47D after 30 min of BMP7 treatment with low endogenous expression of BMP4 (HCC1419, (Figure 2B). At the probe level, both BMP4 and BMP7 SK-BR-3), BMP7 (MDA-MB-231, T-47D) or both hierarchical trees revealed a small subset of gene ele- (HCC1954, MDA-MB-361, ZR-75-30) [12,14] and trea- ments that clustered tightly together (named clusters A ted them with the corresponding BMP ligand (rhBMP4 and B hereafter, Figure 2, blue boxes). However, it is or rhBMP7) and vehicle controls (Figure 1A). Global important to note that the expression of the genes in gene expression levels were analyzed at six different these clusters was not altered in a similar fashion in all time points from 30 min to 24 h in order to reveal the the cell lines; rather, they were upregulated in some cell temporal patterns of transcriptional changes. lines and downregulated in others, showing diverse tem- poral patterns. For both ligands, these clusters appeared Overall transcriptional response to BMP4 and BMP7 to dictate the division of the samples into two major treatment tree branches (Figure 2). Due to the multidimensional nature of our data, we In order to obtain a general view of the cellular func- used three different filtering approaches, each of them tions regulated as a result of BMP4 and BMP7 signaling allowing analysis from a different perspective. Cell line- in breast cancer cell lines, we performed a GO enrich- specific analysis of the expression data evidenced con- ment analysis on the data sets resulting from general fil- siderable variation in the number of differentially tering. As might have been expected, functional expressed probes (DEPs) from one cell line to another categories related to regulation of transcription were (Figure 1B), implicating distinct differences in their tran- among the most highly enriched for both ligands. Addi- scriptional response to BMPs. Further evaluation of tionally, genes involved in organ development were these results revealed that BMP4 treatment resulted in abundantly regulated as a result of stimulation with greater amounts of DEPs than BMP7 (average number either ligand. In this regard, BMP4 seemed more often of DEPs per cell line: 5, 469 versus 3, 898 for BMP4 and to regulate genes involved in skeletal system develop- BMP7, respectively; Figure 1B). This finding could not ment (Table 1), while BMP7, on the other hand, be explained by the differences in cell lines used to appeared to regulate genes involved in epithelial devel- study the two ligands, as a similar outcome was opment, neurogenesis and tube development (Table 2). observed in the three lines treated with both BMP4 and BMP7 (HCC-1954, MDA-MB-361 and ZR-75-30). Time A common synexpression group of genes regulated in point-specific filtering revealed clear temporal variation response to BMP4 and BMP7 signaling in the number of DEPs (Figure 1 C). Generally, there As mentioned above, hierarchical clustering unveiled was a tendency towards a greater amount of DEPs at gene clusters A (BMP4, containing 329 probes) and B later time points. In order to focus our attention espe- (BMP7, 228 probes) with highly correlated expression cially on those genes whose expression was most consis- patterns (Figure 2, blue boxes). Interestingly, of all the tently and extensively affected by BMP4 and BMP7 probes contained in clusters A and B, 210 (154 genes signaling, we performed a general filtering of the with known and unique annotation, named group C Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 5 of 16 http://www.biomedcentral.com/1755-8794/4/80

Figure 2 Unsupervised hierarchical clustering of microarray data. The data sets resulting from general filtering (2, 421 probes for (A) BMP4 and 1, 263 probes for (B) BMP7) were subjected to this analysis. For both ligands, there is an evident cluster of gene elements showing very highly correlated expression patterns throughout the samples (gene clusters A and B, blue boxes). hereafter, Additional file 1) were present in both clusters and ZR-75-30) treated with both ligands revealed ele- and thus represent shared BMP target genes. Direct ments of similarity between BMP4 and BMP7 response comparisons of the expression patterns of group C in the same cell line but high variability between differ- probes in the three cell lines (HCC1954, MDA-MB-361 ent cell lines (Figure 3). For example, these genes were Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 6 of 16 http://www.biomedcentral.com/1755-8794/4/80

Table 1 Enriched gene ontology categories for differentially expressed genes as a result of BMP4 treatment. Category Number of genes p-Value Biological process GO:0006355: regulation of transcription, DNA-dependent 240 0.005 GO:0009888: tissue development 128 0.014 GO:0001501: skeletal system development 53 0.031 GO:0030154: cell differentiation 237 0.041 Molecular function GO:0043565: sequence-specific DNA binding 94 0.043 GO:0003700: sequence-specific DNA binding transcription factor activity 133 0.046 commonly upregulated in MDA-MB-361 while downre- became interested in evaluating whether temporal pat- gulated in HCC1954. The GO enrichment analysis terns of expression and gene function could be related. unveiled 23 enriched biological process terms, of which Therefore, we grouped the clusters from each temporal 21 could be classified into two functional categories, category (early, early-intermediate, late-intermediate, namely development and morphogenesis (16 terms) and and late) and performed GO enrichment analyses on the gene expression (5 terms) (Additional file 2). genes within these groups (Additional file 3). Finally, results from the different cell lines were combined. Temporal patterns of transcriptional response to BMP Although enriched GO terms were not found in every signaling temporal stage of every cell line, many GO terms were Model-based clustering analysis of the expression data enriched in the four temporal categories for BMP4 (Fig- was performed to distinguish clusters of genes with ure 4 C). Of all the functional terms enriched through- similar temporal profiles of expression. With this out the experiments, those related to development were method we identified 12 to 22 probe clusters for each especially abundant in the early and early-intermediate cell line and these clusters could be subsequently classi- phases. Terms connected with regulation of gene fied into four main categories (Tables 3 and 4 and Addi- expression appeared at all times, although they were less tional file 3). Gene elements that were first regulated at abundant at the late stage of 24 h. Metabolism-asso- 30 min or 1 h were classified as early, 3 h or 6 h early- ciated terms were also present throughout the experi- intermediate, and 12 h late-intermediate responders, ment, but were most prominent at late-intermediate regardless of their expression at later time points. Late time points. Signal transduction appeared to be an responders included those probes differentially affected biological process in all but the late-intermedi- expressed exclusively at the 24 h time point. Representa- ate phases; it seemed most profoundly altered during tive examples of clusters in the different temporal cate- early stages. Many terms related to cell proliferation and gories are depicted in Figure 4 A and 4B. DNA repair were enriched among late responder genes, We had already explored the biological functions of the but it should be noted that all these terms emerged genes differentially expressed upon BMP treatment. As from one single cell line, MDA-MB-361. The results we were now able to temporally classify the genes, we from BMP7 data were very limited and may be

Table 2 Enriched gene ontology categories for differentially expressed genes as a result of BMP7 treatment. Category Number of genes p-Value Biological process GO:0030182: neuron differentiation 55 0.024 GO:0048730: epidermis morphogenesis 8 0.025 GO:0051239: regulation of multicellular organismal process 88 0.025 GO:0045944: positive regulation of transcription from RNA polymerase II promoter 42 0.028 GO:0030855: epithelial cell differentiation 24 0.031 GO:0016481: negative regulation of transcription 50 0.033 GO:0035295: tube development 32 0.042 Molecular function GO:0043565: sequence-specific DNA binding 57 0.024 GO:0003705: RNA polymerase II transcription factor activity, enhancer binding 10 0.025 GO:0016564: transcription repressor activity 38 0.038 Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 7 of 16 http://www.biomedcentral.com/1755-8794/4/80

Figure 3 Hierarchical clustering heat map of group C genes with supervised clustering at the sample level. Data from each cell line are grouped, and time points are arranged in temporal order from left to right: 30 min, 1 h, 3 h, 6 h, 12 h and 24 h. summarized as an enrichment of genes involved in regu- ubiquitously regulated by BMP treatments, differentially lation of metabolism and gene expression among the expressed genes resulting from general filtering (1, 678 early-intermediate and late time categories. and 905 for BMP4 and BMP7 experiments, respectively) were ranked according to the number of times a gene Potential novel BMP target genes in breast cancer was up- and downregulated throughout the series of cell One of the main goals of this study was to identify new lines and time points (Tables 5 and 6 and Additional BMP4 and BMP7 target genes relevant in breast cancer. file 4). It is interesting that although the proportion of In order to distinguish those genes most often and up- and down-regulation events was roughly equal Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 8 of 16 http://www.biomedcentral.com/1755-8794/4/80

Table 3 Summary of the temporal clusters obtained from the analysis of BMP4 data. Early Early-intermediate Late-intermediate Late Undetermined cluster genes/ cluster genes/ cluster genes/ cluster genes/ cluster genes/ ID cluster ID cluster ID cluster ID cluster ID cluster HCC1419 2 256 1 273 8 189 6 362 3 178 (14 clusters) 5 148 4 341 11 353 7 152 10 107 14 31 9 224 12 157 13 132 Total 800 654 189 715 554 HCC1954 2 265 4 256 1 158 7 283 (12 clusters) 5 353 9 258 3 419 10 156 6 108 11 244 8 110 12 251 Total 836 1009 577 439 MDA-MB- 6 277 3 155 8 322 1 557 5 289 361 (14 clusters) 13 215 7 177 9 487 2 483 10 362 14 355 12 270 11 554 4 564 Total 847 602 1363 1604 651 SK-BR-3 1 77 2 232 3 235 4 252 (17 clusters) 6 70 10 402 7 365 5 300 11 148 14 383 8 1322 13 567 16 332 9 1104 17 51 12 95 15 508 Total 913 1349 3629 552 ZR-75-30 1 601 9 140 4 2201 8 369 2 10 (18 clusters) 5 107 10 780 14 185 11 688 3 10 12 53 18 436 16 809 6 9 15 227 79 17 1363 13 17 Total 2351 1356 3195 1057 55 Gene clusters were classified in four temporal categories (early, early-intermediate, late-intermediate and late). Clusters with unclear profiles were classified as undetermined. The cluster ID column contains the ID number of the cluster. when considering all the differentially expressed genes, ligands included APOC2 (apolipoprotein C-II) and an induction of gene transcription clearly prevailed over as-yet unnamed gene encoding an uncharacterized pro- inhibition when considering the 100 top-ranked genes tein (C12orf42). DUSP2 (dual specificity phosphatase 2) (75% and 73% of the events for BMP4 and BMP7, and MAP3K5 (mitogen-activated protein kinase kinase respectively). For BMP4, 80 genes were regulated in 10 kinase 5) were highly induced only by BMP4. BMP7, on or more events, while for BMP7, the analogous number the other hand, strongly promoted the expression of of genes was 29 (Tables 5 and 6). Out of the 30 possible genes including PBX1 (pre-B-cell leukemia homeobox 1) regulation events for a single gene (5 cell lines and 6 and ZSCAN4 (zinc finger and SCAN domain containing time points), the actual maximums were 23 events 4), which were not among the genes most intensely (PTPRG; protein tyrosine phosphatase, receptor type, G) regulated by BMP4. for BMP4 and 19 events (GNRHR; gonadotropin-releas- ing hormone receptor) for BMP7 (Tables 5 and 6); all of Discussion these were upregulation events. As expected, several In recent years, it has become increasingly accepted that members of the Id family of inhibitors of DNA binding, the deregulation of mechanisms normally involved in well-known targets of BMPs [27], were strongly induced developmental processes has tumorigenic effects in adult by both BMP ligands. In addition to PTPRG and tissues. One example is the BMP family of growth factors, GNRHR, other genes strongly upregulated by both whose function in cancer physiology has been Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 9 of 16 http://www.biomedcentral.com/1755-8794/4/80

Table 4 Summary of the temporal clusters obtained from the analysis of BMP7 data. Early Early-intermediate Late-intermediate Late Undetermined cluster genes/ cluster genes/ cluster genes/ cluster genes/ cluster genes/ ID cluster ID cluster ID cluster ID cluster ID cluster HCC1954 2 32 4 66 3 226 1 197 6 119 (14 clusters) 5 324 12 113 10 116 13 113 7 113 14 149 835 9 275 11 111 Total 890 328 342 197 232 MDA-MB-231 6 97 4 41 2 175 1 335 (16 clusters) 7 27 9 120 5 321 3 512 10 108 11 101 16 144 8 140 12 125 14 147 13 52 15 85 Total 409 494 640 987 MDA-MB-361 7 240 2 177 1 379 6 252 (12 clusters) 8 193 3 235 5 328 11 258 4 230 9 58 10 336 12 73 Total 691 978 838 252 T-47D 2 767 8 212 1 183 5 1265 14 7 (14 clusters) 3 127 9 594 10 377 6 559 4 38 12 476 7 414 11 163 13 88 Total 1597 1282 560 1824 7 ZR-75-30 3 205 2 169 11 105 14 87 1 8 (22 clusters) 5 286 4 30 12 196 7 16 13 114 6 1544 9 37 15 216 8 35 17 23 20 250 10 1493 21 14 22 36 16 147 18 85 19 336 Total 1107 3839 301 87 98 Gene clusters were classified in four temporal categories (early, early-intermediate, late-intermediate and late). Clusters with unclear profiles were classified as undetermined. The cluster ID column contains the ID number of the cluster. demonstrated in many tumor types, including breast can- ligands, BMP4 and BMP7, on genome-wide gene expres- cer [9-11]. In spite of this, little is known about BMP tar- sion. These two BMPs were selected based on their essen- get genes in the context of tumors. The transcriptional tial role in breast cancer, which we and others have responses of breast cancer cells to BMP signaling have demonstrated in recent years [9-11]. Both ligands are been studied only minimally. More precisely, the effects of highly expressed in primary breast carcinomas as well as BMP2 and BMP7 treatments on transcription in MCF-7 in breast cancer cell lines [12-14,16,17]. BMP7 expression and MDA-MB-468 breast cancer cell lines, respectively, was also shown to be associated with early bone metastasis have been analyzed using cDNA microarrays of limited [15]. Additionally, in vitro studies have implicated BMP4 content (from several hundreds to 14, 500 gene probes) and BMP7 as important regulators of proliferation and [28-30]. In this study, we have therefore set up an experi- migration of breast cancer cells [14,18,20,21,31]. mental procedure to identify potential BMP target genes Our experimental approach allowed multiple types of in breast cancer by studying the effects of two BMP analyses and revealed interesting insights into how the Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 10 of 16 http://www.biomedcentral.com/1755-8794/4/80

Figure 4 Time series analyses. Representative clusters from the four temporal categories are shown for BMP4 (A) and BMP7 (B). For each cluster, the upper figure shows the levels of differential expression through the time series for every probe. The lower chart represents the average value of differential expression for all the probes in the cluster along the time scale. The number of probes in each cluster is indicated under the cell line name. (C) GO analyses of the four temporal categories were performed, and data from the five cell lines were combined. The enriched GO terms for BMP4 are depicted. Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 11 of 16 http://www.biomedcentral.com/1755-8794/4/80

Table 5 Top most regulated genes after BMP4 signaling. # regulated events Gene Id Gene name Gene description Up Down Total ENSG00000144724 PTPRG protein tyrosine phosphatase, receptor type, G 23 0 23 ENSG00000125968 ID1 inhibitor of DNA binding 1, dominant negative helix-loop-helix protein 19 0 19 ENSG00000179088 C12orf42 uncharacterized protein C12orf42 19 0 19 ENSG00000158050 DUSP2 dual specificity phosphatase 2 18 0 18 ENSG00000109163 GNRHR gonadotropin-releasing hormone receptor 18 0 18 ENSG00000234906 APOC2 apolipoprotein C-II 18 0 18 ENSG00000117318 ID3 inhibitor of DNA binding 3, dominant negative helix-loop-helix protein 17 0 17 ENSG00000115738 ID2 inhibitor of DNA binding 2, dominant negative helix-loop-helix protein 16 0 16 ENSG00000172201 ID4 inhibitor of DNA binding 4, dominant negative helix-loop-helix protein 16 0 16 ENSG00000197442 MAP3K5 mitogen-activated protein kinase kinase kinase 5 15 0 15 ENSG00000127129 EDN2 endothelin 2 15 0 15 ENSG00000164850 GPER G protein-coupled estrogen receptor 1 15 0 15 ENSG00000181638 ZFP41 zinc finger protein 41 homolog (mouse) 15 0 15 ENSG00000181626 ANKRD62 ankyrin repeat domain 62 15 0 15 ENSG00000187957 DNER delta/notch-like EGF repeat containing 14 0 14 ENSG00000238243 OR2W3 olfactory receptor, family 2, subfamily W, member 3 14 0 14 ENSG00000164683 HEY1 hairy/enhancer-of-split related with YRPW motif 1 14 0 14 ENSG00000157322 CLEC18A C-type lectin domain family 18, member A 13 1 14 ENSG00000247097 C14orf184 Putative uncharacterized protein C14orf184 8 6 14 ENSG00000212124 TAS2R19 taste receptor, type 2, member 19 9 4 13 ENSG00000186115 CYP4F2 cytochrome P450, family 4, subfamily F, polypeptide 2 6 7 13 ENSG00000176472 ZNF575 zinc finger protein 575 7 6 13 ENSG00000181722 ZBTB20 zinc finger and BTB domain containing 20 7 6 13 ENSG00000163827 LRRC2 leucine rich repeat containing 2 12 0 12 ENSG00000214049 UCA1 urothelial cancer associated 1 12 0 12 ENSG00000132854 KANK4 KN motif and ankyrin repeat domains 4 12 0 12 ENSG00000163749 CCDC158 coiled-coil domain containing 158 12 0 12 ENSG00000100029 PES1 pescadillo homolog 1, containing BRCT domain (zebrafish) 0 12 12 ENSG00000120645 IQSEC3 IQ motif and Sec7 domain 3 0 12 12 ENSG00000122852 SFTPA1 surfactant protein A1 9 3 12 ENSG00000052802 SC4MOL sterol-C4-methyl oxidase-like 9 3 12 ENSG00000197532 OR6Y1 olfactory receptor, family 6, subfamily Y, member 1 8 4 12 ENSG00000078328 RBFOX1 RNA binding protein, fox-1 homolog (C. elegans) 1 8 4 12 ENSG00000185010 F8 coagulation factor VIII, procoagulant component 8 4 12 ENSG00000115756 HPCAL1 hippocalcin-like 1 8 4 12 ENSG00000122859 NEUROG3 neurogenin 3 5 7 12 ENSG00000186810 CXCR3 chemokine (C-X-C motif) receptor 3 6 6 12 ENSG00000168874 ATOH4 atonal homolog 8 (Drosophila) 11 0 11 ENSG00000040731 CDH10 cadherin 10, type 2 (T2-cadherin) 11 0 11 ENSG00000168930 TRIM49 tripartite motif-containing 49 11 0 11 ENSG00000104863 LIN7B lin-7 homolog B (C. elegans) 11 0 11 ENSG00000113391 FAM172A microRNA 2277 11 0 11 ENSG00000162614 NEXN nexilin (F actin binding protein) 11 0 11 ENSG00000120693 SMAD9 SMAD family member 9 11 0 11 ENSG00000176907 C8orf4 Uncharacterized protein C8orf4 (Thyroid cancer protein 1)(TC-1) 10 1 11 ENSG00000085224 ATRX alpha thalassemia/mental retardation syndrome X-linked 10 1 11 ENSG00000163743 RCHY1 ring finger and CHY zinc finger domain containing 1 9 2 11 ENSG00000006007 GDE1 glycerophosphodiester phosphodiesterase 1 9 2 11 ENSG00000114850 SSR3 signal sequence receptor, gamma (translocon-associated protein gamma) 9 2 11 ENSG00000109472 CPE carboxypeptidase E 9 2 11 Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 12 of 16 http://www.biomedcentral.com/1755-8794/4/80

Table 5 Top most regulated genes after BMP4 signaling. (Continued) ENSG00000106608 URGCP upregulator of cell proliferation 2 9 11 ENSG00000064201 TSPAN32 tetraspanin 32 8 3 11 ENSG00000212128 TAS2R13 taste receptor, type 2, member 13 8 3 11 ENSG00000092871 RFFL ring finger and FYVE-like domain containing 1 7 4 11 ENSG00000139880 CDH24 cadherin 24, type 2 4 7 11 ENSG00000039319 ZFYVE16 zinc finger, FYVE domain containing 16 4 7 11 ENSG00000206052 DOK6 docking protein 6 6 5 11 ENSG00000050165 DKK3 dickkopf homolog 3 (Xenopus laevis) 5 6 11 ENSG00000115844 DLX2 distal-less homeobox 2 10 0 10 ENSG00000178343 SHISA3 shisa homolog 3 (Xenopus laevis) 10 0 10 ENSG00000123329 ARHGAP9 Rho GTPase activating protein 9 10 0 10 ENSG00000145287 PLAC8 placenta-specific 8 10 0 10 ENSG00000187634 SAMD11 sterile alpha motif domain containing 11 10 0 10 ENSG00000167962 ZNF598 zinc finger protein 598 10 0 10 ENSG00000003509 C2orf56 Protein midA homolog, mitochondrial Precursor 8 2 10 ENSG00000143153 ATP1B1 ATPase, Na+/K+ transporting, beta 1 polypeptide 8 2 10 ENSG00000189079 ARID2 AT rich interactive domain 2 (ARID, RFX-like) 2 8 10 ENSG00000162706 CADM3 cell adhesion molecule 3 7 3 10 ENSG00000147488 ST18 suppression of tumorigenicity 18 (breast carcinoma) (zinc finger protein) 7 3 10 ENSG00000182175 RGMA RGM domain family, member A 7 3 10 ENSG00000163623 NKX6-1 NK6 homeobox 1 3 7 10 ENSG00000144559 C3orf31 MMP37-like protein, mitochondrial Precursor 6 4 10 ENSG00000153002 CPB1 carboxypeptidase B1 (tissue) 6 4 10 ENSG00000181965 NEUROG1 neurogenin 1 6 4 10 ENSG00000171564 FGB fibrinogen beta chain 6 4 10 ENSG00000132612 VPS4A vacuolar protein sorting 4 homolog A (S. cerevisiae) 6 4 10 ENSG00000183023 SLC8A1 solute carrier family 8 (sodium/calcium exchanger), member 1 4 6 10 ENSG00000166748 AGBL1 ATP/GTP binding protein-like 1 5 5 10 ENSG00000204882 GPR20 G protein-coupled receptor 20 5 5 10 ENSG00000159216 RUNX1 runt-related transcription factor 1 5 5 10 Differentially expressed genes resulting from the general filtering of BMP4 data (1, 678) were ranked according to the number of times a gene was up- and downregulated throughout the series of cell lines and time points. Only those genes with total rank value of at least 10 are listed. stimulation of BMP4 and BMP7 signaling influences the in a given cell. We have previously reported that all six transcriptome of breast cancer cell lines. First of all, we BMP specific receptors (ACVR1, BMPR1A, BMPR1B, showed relatively high numbers of DEPs as a result of ACVR2A,ACVR2B,andBMPR2)areuniformly BMP treatments, indicating a strong impact of BMP4 expressed among the breast cancer cell lines studied and BMP7 on the cell lines studied. Moreover, the tran- here [12]. Similarly, we have shown that SMAD4 is scriptional response to BMP4 was of a clearly higher expressed and that phosphorylation of SMAD-1/5/8 is magnitude than that induced by BMP7. Another aspect induced in these cell lines after BMP7 and BMP4 treat- of the study was the opportunity to compare the effects ment [14,21]. Taken together, the expression profiles of of BMP signaling between cell lines. Interestingly, clear BMP specific receptors or the mediators of the canoni- differences were seen in the amounts of DEPs, as well as cal intracellular pathway do not seem to have a major in their expression patterns, as revealed by hierarchical role in explaining the different transcriptional responses clustering. BMP signaling pathways are regulated in a in the breast cancer cells. Nevertheless, due to the com- very complex manner and at many different levels, from plexity of BMP signaling regulation, it is easy to under- the availability of BMP receptors, BMP ligands and stand that different cell lines may have different BMP antagonists in the extracellular compartment to transcriptional responses to BMP stimulation. This thepresenceorabsenceofvarious intracellular signal observation highlights the importance of testing multiple mediators and transcriptional co-activators or co-repres- cell lines when studying BMP signaling in cancer. An sors [32,33]. Therefore, multiple factors influence the additional finding that could be inferred from our data outcome of BMP signaling on the transcriptional level is that induction of gene transcription, compared with Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 13 of 16 http://www.biomedcentral.com/1755-8794/4/80

Table 6 Top most regulated genes after BMP7 signaling. # regulated events Gene Id Gene name Gene description Up Down Total ENSG00000109163 GNRHR gonadotropin-releasing hormone receptor 19 0 19 ENSG00000157322 CLEC18A C-type lectin domain family 18, member A 16 1 17 ENSG00000179088 C12orf42 Uncharacterized protein C12orf42 15 0 15 ENSG00000123329 ARHGAP9 Rho GTPase activating protein 9 14 0 14 ENSG00000144724 PTPRG protein tyrosine phosphatase, receptor type, G 14 0 14 ENSG00000185630 PBX1 pre-B-cell leukemia homeobox 1 13 0 13 ENSG00000114850 SSR3 signal sequence receptor, gamma (translocon-associated protein gamma) 9 4 13 ENSG00000186153 WWOX WW domain containing oxidoreductase 5 8 13 ENSG00000234906 APOC2 apolipoprotein C-II 12 0 12 ENSG00000180532 ZSCAN4 zinc finger and SCAN domain containing 4 12 0 12 ENSG00000144711 IQSEC1 IQ motif and Sec7 domain 1 12 0 12 ENSG00000138821 SLC39A8 solute carrier family 39 (zinc transporter), member 8 11 1 12 ENSG00000165995 CACNB2 calcium channel, voltage-dependent, beta 2 subunit 3 9 12 ENSG00000122859 NEUROG3 neurogenin 3 8 4 12 ENSG00000212128 TAS2R13 taste receptor, type 2, member 13 6 6 12 ENSG00000184999 SLC22A10 solute carrier family 22, member 10 11 0 11 ENSG00000183914 DNAH2 dynein, axonemal, heavy chain 2 11 0 11 ENSG00000125968 ID1 inhibitor of DNA binding 1, dominant negative helix-loop-helix protein 11 0 11 ENSG00000167962 ZNF598 zinc finger protein 598 10 1 11 ENSG00000181722 ZBTB20 zinc finger and BTB domain containing 20 3 8 11 ENSG00000091482 SMPX small muscle protein, X-linked 10 0 10 ENSG00000117318 ID3 inhibitor of DNA binding 3, dominant negative helix-loop-helix protein 10 0 10 ENSG00000163694 RBM47 RNA binding motif protein 47 10 0 10 ENSG00000006007 GDE1 glycerophosphodiester phosphodiesterase 1 9 1 10 ENSG00000164104 HMGB2 high-mobility group box 2 8 2 10 ENSG00000250589 DUX4 double homeobox 4 2 8 10 ENSG00000130559 CAMSAP1 calmodulin regulated spectrin-associated protein 1 3 7 10 ENSG00000166501 PRKCB protein kinase C, beta 4 6 10 ENSG00000186115 CYP4F2 cytochrome P450, family 4, subfamily F, polypeptide 2 4 6 10 Differentially expressed genes resulting from the general filtering of BMP7 data (905) were ranked according to the number of times a gene was up- and downregulated throughout the series of cell lines and time points. Only those genes with total rank value of at least 10 are listed. inhibition, was the common response among those Synexpression groups are synchronously coexpressed genes most frequently regulated by BMP4 and BMP7 in gene sets, particularly apparent during embryonic devel- breast cancer cells. Likewise, previous microarray-based opment and in the response of cells to hormones and transcriptomic analyses of TGF-b and BMP have shown growth factors [38,39]. Our analyses unveiled that treat- that induction of gene expression is the predominant ment of breast cancer cells with either BMP4 or BMP7 response of mammalian cells to stimulation by these resulted in the coordinated expression of a group of growth factors [28,30,33-35]. genes (clusters A and B, respectively). Most interest- After BMP4 and BMP7 stimulation, the microarray ingly, a considerable number of the genes in these two analyses identified a large number of differentially synexpression groups were common for the two ligands expressed genes in our panel of cell lines. To explore (group C). Moreover, our data indicated that treatment the biological functions of these genes, GO enrichment of a cell line with either BMP4 or BMP7 results in simi- analyses were performed. These revealed very similar lar transcriptional responses of group C genes. We results for both BMP ligands, namely, regulation of tran- therefore hypothesize that group C represents molecular scription and developmental processes. It seems, there- responses shared by the BMP4 and BMP7 signaling fore, that the functions most prevalently influenced by pathways. This finding prompted us to ask what func- BMP signaling in breast cancer cells do not differ tions these common genes fulfill in the cell. GO enrich- remarkably from conventional roles that BMPs possess ment analysis of the genes in group C revealed that during development [36,37]. these genes are involved in two main biological Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 14 of 16 http://www.biomedcentral.com/1755-8794/4/80

processes, regulation of gene expression and regulation various aspects of normal and malignant breast biology of development and morphogenesis. These results sup- [41]. Others are newly linked, in this work, to BMP sig- port the notion that genes known to regulate develop- naling, and some of these genes have interesting con- ment also have functions that are important for the nections with tumor biology, such as PTPRG or DUSP2. maintenance of cancer cells. A positive feedback regulation where BMP treatment We also studied the temporal patterns of the tran- leads to increased expression of BMP antagonists is scriptional response after BMP treatment. The number known to exist. In our study, no consistent expression of DEPs showed a tendency to increase with time, a changes were observed for any of the known BMP trend previously noticed in transcriptome analysis of antagonists, such as noggin, gremlin, sclerostin and fol- TGF-b family members in murine mammary epithelial listatin. Previous studies have shown a wide time win- cells and in breast cancer cells [29,35]. The DEPs could dowintheinductionofe.g.nogginexpressionin be grouped according to their temporal pattern of different tissues after BMP treatment, ranging from 1 to expression, varying from early to late responders. These 48 hours [42,43]. Thus it is possible that the feedback temporal clusters were found in every cell line, and effect in the breast cancer cells was not evident at time some of them even contained over a thousand gene ele- points analyzed here. ments. The next logical step was to explore whether Protein tyrosine phosphatases (PTPs) are key regula- there was a time-dependent shift in the distribution of tors of the cellular protein phosphorylation balance, cri- gene functions. Although GO enrichment results were tical in the control of a wide spectrum of physiological not obtained for all the probe clusters of all the cell processessuchascellproliferation, differentiation, lines, interesting features could be identified, especially transformation, transport and locomotion. Subsequently, in the case of the BMP4 data. Transcriptional regulation aberrations in phosphorylation processes play a major in the first 6 hours concentrated most notably on genes role in the pathogenesis of numerous diseases, including involved in developmental processes, metabolic pro- cancer [44,45]. PTPRG is a receptor-type PTP impli- cesses, gene expression and signal transduction. Gene cated as a candidate tumor suppressor gene in several expression was also well-represented after 12 hours, types of tumors, including breast cancer [46,47]. In while metabolism became by far the most prominent MCF-7 breast cancer cells, PTPRG inhibits proliferation function at this time point. Most interestingly, 24 hours and anchorage-independent growth and reduces tumor after BMP4 stimulation there was an evident overrepre- formation in a xenograft model [47,48]. Delayed cell sentation of genes involved in cell proliferation, cycle re-entry by increasing the level of cell cycle regula- although this phenomenon was observed exclusively in tors p21 and p27 could explain the inhibitory effect of MDA-MB-361 cells. All in all, the enriched biological PTPRG on cell growth [47]. Based on the above, upre- functions indeed fluctuated in time and in a logical gulation of PTPRG in BMP-stimulated cancer cells sequence, with regulation of gene expression and signal could contribute to the observed BMP-induced antipro- transduction leading to changes in metabolism and liferative effect [14,21]. DUSP2 also belongs to the PTP finally to regulation of cell proliferation, a phenotype family of phosphatases. It is a mitogen-activated protein relevant for cancer cell physiology. The fact that we did kinase (MAPK) phosphatase (MKP) that dephosphory- not see enrichment of cell proliferation-associated func- lates both threonine and tyrosine residues within target tions in more than one cell line could be due to differ- MAPKs leading to their deactivation. MAPK signaling ences in the speed of BMP signaling in different cell controls cellular processes such as proliferation, differ- lines. Even though a longer experiment certainly could entiation, migration and apoptosis [48]. Therefore, have clarified this issue, we concentrated our analysis on abnormal MKP activity, and hence anomalous MAPK the first 24 hours after BMP treatment because we were signaling, has important consequences for processes cri- interested primarily in the identification of BMP target tical to the development and progression of human can- genes. cer. The role of DUSP2 in cancer has been examined in As mentioned, one of the main goals of this study was only a few studies, and data are controversial. Overex- to identify potential novel gene targets of BMP signaling pression of DUSP2 expression was found in 37 of 39 relevant in breast cancer. We provided lists of candidate malignant effusions from serous ovarian carcinoma genes that are strongly and rather uniformly regulated patients and was associated with poor survival [49]. By by BMP4 or BMP7 throughout the cell lines and time contrast, decreased DUSP2 transcript levels were points. Some of them, such as members of the Id family reported in cancerous breast, colon, lung, ovary, kidney of inhibitors of DNA binding, are well-known BMP tar- and prostate tissues, and reduced DUSP2 protein levels get genes [27,29]. Id proteins are transcription factors were observed in cervical and colon cancer [50]. Addi- that regulate cell growth and differentiation [40], and all tionally, DUSP2 suppression was associated with tumori- four members of the protein family play crucial roles in genesis and malignancy in colon cancer, and DUSP2 Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 15 of 16 http://www.biomedcentral.com/1755-8794/4/80

overexpression induced apoptosis and inhibited tumor References in vitro 1. Schmierer B, Hill CS: TGFbeta-SMAD signal transduction: molecular growth in HeLa cells and in xenograft models specificity and functional flexibility. Nat Rev Mol Cell Biol 2007, 8:970-982. [51]. 2. Nohe A, Keating E, Knaus P, Petersen NO: Signal transduction of bone morphogenetic protein receptors. Cell Signal 2004, 16:291-299. Conclusions 3. Derynck R, Zhang YE: Smad-dependent and Smad-independent pathways in TGF-beta family signalling. Nature 2003, 425:577-584. All in all, we show that BMP4 and BMP7 have strong 4. Attisano L, Labbe E: TGFbeta and Wnt pathway cross-talk. Cancer effects on gene expression in breast cancer cells, and Metastasis Rev 2004, 23:53-61. that this transcriptional response and its functional out- 5. Miyazono K, Maeda S, Imamura T: BMP receptor signaling: transcriptional targets, regulation of signals, and signaling cross-talk. Cytokine Growth come follow a temporal sequence. Our data support the Factor Rev 2005, 16:251-263. existence of a synexpression group of regulated genes 6. Herpin A, Cunningham C: Cross-talk between the bone morphogenetic that represent the core molecular responses shared by protein pathway and other major signaling pathways results in tightly regulated cell-specific outcomes. Febs J 2007, 274:2977-2985. BMP4 and BMP7 signaling pathways. Additionally, we 7. Wozney JM, Rosen V, Celeste AJ, Mitsock LM, Whitters MJ, Kriz RW, provide a list of potential novel BMP target genes rele- Hewick RM, Wang EA: Novel regulators of bone formation: molecular vant in breast cancer. clones and activities. Science 1988, 242:1528-1534. 8. Hogan BL: Bone morphogenetic proteins: multifunctional regulators of vertebrate development. Genes Dev 1996, 10:1580-1594. Additional material 9. Thawani JP, Wang AC, Than KD, Lin CY, La Marca F, Park P: Bone morphogenetic proteins and cancer: review of the literature. Neurosurgery 2010, 66:233-246, discussion 246. Additional file 1: List of probes in group C. The expression values are 10. Singh A, Morris RJ: The Yin and Yang of bone morphogenetic proteins in represented as log2 ratios. Data was colored according to the cancer. Cytokine Growth Factor Rev 2010, 21:299-313. accompanying color key. The data sheet contains 210 probes. 11. Alarmo EL, Kallioniemi A: Bone morphogenetic proteins in breast cancer: Additional file 2: GO enrichment results for group C genes. dual role in tumourigenesis? Endocr Relat Cancer 2010, 17:R123-139. 12. Alarmo EL, Kuukasjarvi T, Karhu R, Kallioniemi A: A comprehensive Additional file 3: Model-based clustering results and GO expression survey of bone morphogenetic proteins in breast cancer enrichment. The probe clusters obtained are represented as well as the highlights the importance of BMP4 and BMP7. Breast Cancer Res Treat results of the GO enrichment analyses. GO terms are colored according 2007, 103:239-246. to the accompanying color key. 13. Schwalbe M, Sanger J, Eggers R, Naumann A, Schmidt A, Hoffken K, Additional file 4: Most commonly regulated genes with rank 6 or Clement JH: Differential expression and regulation of bone higher. The expression values are represented as log2 ratios. Data was morphogenetic protein 7 in breast cancer. Int J Oncol 2003, 23:89-95. colored according to the accompanying color key. The data sheets 14. Ketolainen JM, Alarmo EL, Tuominen VJ, Kallioniemi A: Parallel inhibition of contain 608 and 281 genes for BMP4 and BMP7, respectively. cell growth and induction of cell migration and invasion in breast cancer cells by bone morphogenetic protein 4. Breast Cancer Res Treat 2010. 15. Alarmo EL, Korhonen T, Kuukasjarvi T, Huhtala H, Holli K, Kallioniemi A: Acknowledgements Bone morphogenetic protein 7 expression associates with bone The authors greatly appreciate the skillful assistance of Ms. Kati Rouhento. metastasis in breast carcinomas. Ann Oncol 2008, 19:308-314. This study was partly supported by grants from the Competitive Research 16. Alarmo EL, Rauta J, Kauraniemi P, Karhu R, Kuukasjarvi T, Kallioniemi A: Bone Funding of the Tampere University Hospital, the Sigrid Jusélius Foundation, morphogenetic protein 7 is widely overexpressed in primary breast the Finnish Cancer Organizations, the Academy of Finland (project: 125826) cancer. Genes Cancer 2006, 45:411-419. and Biocentrum Helsinki. 17. Schwaninger R, Rentsch CA, Wetterwald A, van der Horst G, van Bezooijen RL, van der Pluijm G, Lowik CW, Ackermann K, Pyerin W, Author details Hamdy FC, et al: Lack of noggin expression by cancer cells is a 1Laboratory of Cancer Genetics, Institute of Biomedical Technology, determinant of the osteoblast response in bone metastases. Am J Pathol University of Tampere and Centre for Laboratory Medicine, Tampere 2007, 170:160-175. University Hospital, Finland. 2Computational Systems Biology Laboratory, 18. Buijs JT, Henriquez NV, van Overveld PG, van der Horst G, Que I, Genome-Scale Biology Research Program, Institute of Biomedicine, University Schwaninger R, Rentsch C, Ten Dijke P, Cleton-Jansen AM, Driouch K, et al: of Helsinki, Finland. Bone morphogenetic protein 7 in the development and treatment of bone metastases from breast cancer. Cancer Res 2007, 67:8742-8751. Authors’ contributions 19. Montesano R: Bone morphogenetic protein-4 abrogates lumen formation ARM participated in the design of the study, carried out the interpretation of by mammary epithelial cells and promotes invasive growth. Biochem the data analyses and drafted the manuscript. ELA was responsible for Biophys Res Commun 2007, 353:817-822. conception and design of the project and participated in writing the 20. Shon SK, Kim A, Kim JY, Kim KI, Yang Y, Lim JS: Bone morphogenetic manuscript. LS performed computational data analyses and participated in protein-4 induced by NDRG2 expression inhibits MMP-9 activity in writing the manuscript. JK performed the microarray hybridizations. KN breast cancer cells. Biochem Biophys Res Commun 2009, 385:198-203. performed computational data analyses. SH supervised and coordinated the 21. Alarmo EL, Parssinen J, Ketolainen JM, Savinainen K, Karhu R, Kallioniemi A: computational data analyses and participated in writing the manuscript. AK BMP7 influences proliferation, migration, and invasion of breast cancer was responsible for conception and coordination of the project and cells. Cancer Lett 2009, 275:35-43. participated in writing the manuscript. All authors read and approved the 22. csbl.go: Semantic similarity based on Gene Ontology. [http://csbi.ltdk. final manuscript. helsinki.fi/csbl.go]. 23. Fraley C, Raftery AE: Model based clustering, discriminant analysis, and Competing interests density estimation. J Am Stat Assoc 2002, 97:611-631. The authors declare that they have no competing interests. 24. Yeung KY, Fraley C, Murua A, Raftery AE, Ruzzo WL: Model-based clustering and data transformations for gene expression data. Received: 12 September 2011 Accepted: 25 November 2011 Bioinformatics 2001, 17:977-987. Published: 25 November 2011 25. Ovaska K, Laakso M, Haapa-Paananen S, Louhimo R, Chen P, Aittomaki V, Valo E, Nunez-Fontarnau J, Rantanen V, Karinen S, et al: Large-scale data Rodriguez-Martinez et al. BMC Medical Genomics 2011, 4:80 Page 16 of 16 http://www.biomedcentral.com/1755-8794/4/80

integration framework provides a comprehensive view on glioblastoma (PTPgamma) in human breast cancer cell line MCF-7. Oncogene 2004, multiforme. Genome Med 2010, 2:65. 23:1256-1262. 26. R_Development_Core_Team: R: A language and environment for 49. Chang L, Karin M: Mammalian MAP kinase signalling cascades. Nature statistical computing. Vienna, Austria: R Foundation for Statistical 2001, 410:37-40. Computing; 2011. 50. Givant-Horwitz V, Davidson B, Goderstad JM, Nesland JM, Trope CG, 27. Hollnagel A, Oehlmann V, Heymer J, Ruther U, Nordheim A: Id genes are Reich R: The PAC-1 dual specificity phosphatase predicts poor outcome direct targets of bone morphogenetic protein induction in embryonic in serous ovarian carcinoma. Gynecol Oncol 2004, 93:517-523. stem cells. J Biol Chem 1999, 274:19838-19845. 51. Lin SC, Chien CW, Lee JC, Yeh YC, Hsu KF, Lai YY, Lin SC, Tsai SJ: 28. Kowanetz M, Valcourt U, Bergstrom R, Heldin CH, Moustakas A: Id2 and Id3 Suppression of dual-specificity phosphatase-2 by hypoxia increases define the potency of cell proliferation and differentiation responses to chemoresistance and malignancy in human cancer cells. J Clin Invest transforming growth factor beta and bone morphogenetic protein. Mol 2011, 121:1905-1916. Cell Biol 2004, 24:4241-4254. 29. Clement JH, Marr N, Meissner A, Schwalbe M, Sebald W, Kliche KO, Pre-publication history Hoffken K, Wolfl S: Bone morphogenetic protein 2 (BMP-2) induces The pre-publication history for this paper can be accessed here: sequential changes of Id gene expression in the breast cancer cell line http://www.biomedcentral.com/1755-8794/4/80/prepub MCF-7. J Cancer Res Clin Oncol 2000, 126:271-279. 30. Steinert S, Kroll TC, Taubert I, Pusch L, Hortschansky P, Hoffken K, Wolfl S, doi:10.1186/1755-8794-4-80 Clement JH: Differential expression of cancer-related genes by single Cite this article as: Rodriguez-Martinez et al.: Analysis of BMP4 and and permanent exposure to bone morphogenetic protein 2. J Cancer Res BMP7 signaling in breast cancer cells unveils time-dependent Clin Oncol 2008, 134:1237-1245. transcription patterns and highlights a common synexpression group of 31. Takahashi M, Otsuka F, Miyoshi T, Otani H, Goto J, Yamashita M, Ogura T, genes. BMC Medical Genomics 2011 4:80. Makino H, Doihara H: Bone morphogenetic protein 6 (BMP6) and BMP7 inhibit estrogen-induced proliferation of breast cancer cells by suppressing p38 mitogen-activated protein kinase activation. J Endocrinol 2008, 199:445-455. 32. Zeng S, Chen J, Shen H: Controlling of bone morphogenetic protein signaling. Cell Signal 2010, 22:888-893. 33. Massague J, Seoane J, Wotton D: Smad transcription factors. Genes Dev 2005, 19:2783-2810. 34. de Jong DS, Steegenga WT, Hendriks JM, van Zoelen EJ, Olijve W, Dechering KJ: Regulation of Notch signaling genes during BMP2-induced differentiation of osteoblast precursor cells. Biochem Biophys Res Commun 2004, 320:100-107. 35. Valcourt U, Kowanetz M, Niimi H, Heldin CH, Moustakas A: TGF-beta and the Smad signaling pathway support transcriptomic reprogramming during epithelial-mesenchymal cell transition. Mol Biol Cell 2005, 16:1987-2002. 36. Chen D, Zhao M, Mundy GR: Bone morphogenetic proteins. Growth Factors 2004, 22:233-241. 37. Gazzerro E, Canalis E: Bone morphogenetic proteins and their antagonists. Rev Endocr Metab Disord 2006, 7:51-65. 38. Baldessari D, Shin Y, Krebs O, Konig R, Koide T, Vinayagam A, Fenger U, Mochii M, Terasaka C, Kitayama A, et al: Global gene expression profiling and cluster analysis in Xenopus laevis. Mech Dev 2005, 122:441-475. 39. Karaulanov E, Knochel W, Niehrs C: Transcriptional regulation of BMP4 synexpression in transgenic Xenopus. Embo J 2004, 23:844-856. 40. Norton JD, Deed RW, Craggs G, Sablitzky F: Id helix-loop-helix proteins in cell growth and differentiation. Trends Cell Biol 1998, 8:58-65. 41. de Candia P, Benera R, Solit DB: A role for Id proteins in mammary gland physiology and tumorigenesis. Adv Cancer Res 2004, 92:81-94. 42. Stottmann RW, Anderson RM, Klingensmith J: The BMP antagonists chordin and noggin have essential but redundant roles in mouse mandibular outgrowth. Dev Biol 2001, 240:457-473. 43. Haudenschild DR, Palmer SM, Moseley TA, You Z, Reddi AH: Bone morphogenetic protein (BMP)-6 signaling and BMP antagonist noggin in prostate cancer. Cancer Res 2004, 64:8276-8284. 44. Yi T, Lindner D: The role and target potential of protein tyrosine phosphatases in cancer. Curr Oncol Rep 2008, 10:114-121. 45. Julien SG, Dube N, Hardy S, Tremblay ML: Inside the human cancer Submit your next manuscript to BioMed Central tyrosine phosphatome. Nat Rev Cancer 2011, 11:35-49. and take full advantage of: 46. LaForgia S, Morse B, Levy J, Barnea G, Cannizzaro LA, Li F, Nowell PC, Boghosian-Sell L, Glick J, Weston A, et al: Receptor protein-tyrosine phosphatase gamma is a candidate tumor suppressor gene at human • Convenient online submission region 3p21. Proc Natl Acad Sci USA 1991, 88:5036-5040. • Thorough peer review 47. Shu ST, Sugimoto Y, Liu S, Chang HL, Ye W, Wang LS, Huang YW, Yan P, • No space constraints or color figure charges Lin YC: Function and regulatory mechanisms of the candidate tumor suppressor receptor protein tyrosine phosphatase gamma (PTPRG) in • Immediate publication on acceptance breast cancer cells. Anticancer Res 2010, 30:1937-1946. • Inclusion in PubMed, CAS, Scopus and Google Scholar 48. Liu S, Sugimoto Y, Sorio C, Tecchio C, Lin YC: Function analysis of • Research which is freely available for redistribution estrogenically regulated protein tyrosine phosphatase gamma

Submit your manuscript at www.biomedcentral.com/submit